Author: Gregor J. Rothfuss

Cell Compartmentalization

When compartmentalization was thought of as a singular feature of eukaryotes, experts were often forced to speculate about how it came about, what the biophysical constraints were, and what selective advantages it might have. “That’s where these prokaryotes become really interesting. If they show some features that are even mildly similar to what we see in eukaryotes, it allows us to broaden the question and attack from a different angle: Under what conditions might compartmentation provide some benefit? Or is it just the case that there’s no benefit whatsoever?” The bacterial cases “suggest that there are multiple ways to do this, and that there could be a strong evolutionary advantage to doing so.” That certainly seems to have been the case with energy production: The independent evolution of anammoxosomes in some kinds of bacteria and mitochondria in eukaryotes signify that “the compartmentalization of energy metabolism is beneficial to the cell. You see a trend in both prokaryotes and eukaryotes to compartmentalize certain traits or certain functions in order to better control them.”

2021-01-10: Biomolecular condensates are an emerging type of membrane-less organelle, and adding a lot to our understanding of cell compartmentalization.

Inside cells, droplets of biomolecules called condensates merge, divide and dissolve. Their dance may regulate vital processes. Biomolecular Condensates may explain the speed of many cellular processes. “The key thing about a condensate — it’s not like a factory; it’s more like a flash mob. You turn on the radio, and everyone comes together, and then you turn it off and everyone disappears”


2022-11-02: Condensates are getting more refined experiments to figure out what’s going on

After an initial rush to document the phenomenon in every nook and cranny of the cell, scientists are beginning to ask more detailed questions. They want to know what these globules are doing, how they form and, importantly, how to prove that these biomolecular condensates are really as widespread and essential to the cell as many reports have claimed. Researchers are also responding to critics who have questioned the accuracy of some descriptions of phase separation in cells, arguing that other forces besides phase separation could have created droplets.
Inside the condensates, enzymatic reactions were 36x faster. Condensates gave the process extra structure: they helped to organize the enzymes spatially, providing a molecular ‘scaffold’ so that they could more easily partner with their reactants. “You get this combined effect of increasing efficiency and increasing concentration”
Some in the community have sought to inject precision into the field and guide researchers in finding out whether a blob forms through phase separation or in some other way. Despite the debates in academia, drug hunters are embracing the concept. Most companies interested in phase separation are prioritizing drug development for cancer and neurological disorders, 2 disease classes frequently linked to condensates that have gone awry.

DNN parallelization strategies

Traditional approaches to training exploit either data parallelism (dividing up the training samples), model parallelism (dividing up the model parameters), or expert-designed hybrids for particular situations. FlexFlow encompasses both of these in its sample (data parallelism), and parameter (model parallelism) dimensions, and also adds an operator dimension (more model parallelism) describing how operators within a DNN should be parallelized, and an attribute dimension with defines how different attributes within a sample should be partitioned (e.g. height and width of an image).

Do Traditions still matter?

1 way to ensure survival is high-fidelity adherence to traditions + ensuring that the inherited ancestral environment/context is more or less maintained. Adhering to ancient traditions when the context is rapidly changing is a recipe for disaster. No point in mastering seal-hunting if there ain’t no more seals. No point in mastering the manners of being a courtier if there ain’t no more royal court. Etc. And the problem is that, in the modern world, we can’t simply all mutually agree to stop changing our context so that our traditions will continue to function as before because it is no longer under our control.

Amazon Surveillance State

Amazon wants you to be part of its dish network. Yes, it’s a play on words (and not a good one!). This network springs from Amazon’s Ring doorbell — the doorbell with a camera inside and a cozy relationship with law enforcement! What are your neighbors and strangers up to? Give the dirt to law enforcement and trust their better judgment!

Good times await those who find themselves looking dark or suspicious (but also suspicious because they’re dark) in front of a Ring doorbell. Have you ever wanted to be an internet celebrity, with or without your permission? Ring has you covered.

Solar System Orbit Map

This week’s map shows the orbits of 18k asteroids in the solar system. This includes everything we know of that’s over 10km in diameter – 10k asteroids – as well as 8k randomized objects of unknown size. This map shows each asteroid at its exact position on New Years’ Eve 1999. All of the data for this map is shared by NASA and open to the public. However, the data is stored in several different databases so I had to do a decent amount of data cleaning. I’ve explained all of the steps in detail in my open-source code and tutorial, so I’ll just include a sketch of the process here in this blog post.

Fourier Transform

the Fourier transform tells you how much of each ingredient “note” (sine wave or circle) contributes to the overall wave. Here’s why Fourier’s trick is useful. Imagine you were talking to your friend over the phone and you wanted to get them to draw this squarish wave. The tedious way to do this would be to read out a long list of numbers that represent the height of the wave at every instant in time. With all these numbers, your friend could patiently stitch together the original wave. This is essentially how old audio formats like WAV files worked. But if your friend knew Fourier’s trick, you could do something pretty slick: You could just tell them a handful of numbers—the sizes of the different circles in the picture above. They can then use this circle picture to reconstruct the original wave.

2022-11-14: The nuclear origins of Fast Fourier Transforms

And this trick works even if the signal is composed of a bunch of different frequencies. If the sine waves frequency is one of the components of the signal it will correlate with the signal producing a non-0 area. And the size of this area tells you the relative amplitude of that frequency sine wave in the signal. Repeat this process for all frequencies of sine waves and you get the frequency spectrum. Essentially which frequencies are present and in what proportions. If the signal is a cosine wave, then even if you multiply it by a sine wave of the exact same frequency, the area under the curve will be 0. For each frequency, we need to multiply by a sine wave and a cosine wave and find the amplitudes for each. The ratio of these amplitudes indicates the phase of the signal that is how much it’s shifted to the left or to the right. You can use Euler’s formula so you only need to multiply your signal by one exponential term. Then the real part of the sum is the cosine amplitude and the imaginary part is the sine amplitude.

Privacy economics

the major tech companies are much less of a threat to our actual privacy than is typically assumed. For most people, gossip from friends, relatives, colleagues, and acquaintances is a bigger privacy risk than is information garnered on-line. Gossip is an age-old problem, and still today many of the biggest privacy harms come through very traditional channels. Privacy is a real issue, but to the extent it can be fixed, most of that needs to happen outside of the major tech companies. Most of what is written about tech and privacy is simply steering us down the wrong track.

Tupperware

While Tupperware shares many common features with other cluster management systems, such as Kubernetes and Mesos, it distinguishes itself in the following areas: Seamless support for stateful services. A single control plane managing servers across data centers to help automate intent-based container deployment, cluster decommission, and maintenance. Transparent sharding of the control plane to scale out. An elastic compute approach to shift capacity among services in real time.

Efficient, reliable cluster management at scale with Tupperware